historical seismograms for unravelling a mysterious ... · seismic stations in the world. as is...
TRANSCRIPT
Geophys. J. Int. (2010) 183, 358–374 doi: 10.1111/j.1365-246X.2010.04731.x
GJI
Sei
smol
ogy
Historical seismograms for unravelling a mysterious earthquake:The 1907 Sumatra Earthquake
Hiroo Kanamori,1 Luis Rivera2 and William H. K. Lee3
1Seismological Laboratory, California Institute of Technology, Pasadena, CA 91125, USA. E-mail: [email protected] de Physique du Globe de Strasbourg; Universite de Strasbourg-CNRS, Strasbourg, France3862 Richardson Court, Palo Alto, CA 94303, USA
Accepted 2010 July 8. Received 2010 June 29; in original form 2009 November 16
S U M M A R YHistory of instrumental seismology is short. Seismograms are available only for a little morethan 100 years; high-quality seismograms are available only for the last 50 years and theseismological database is very limited in time. To extend the database, seismograms of oldevents are of vital importance. Many unusual earthquakes are known to have occurred, buttheir seismological characteristics are poorly known. The 1907 Sumatra earthquake is one ofthem (1907 January 4, M = 7.6). Gutenberg and Richter located this event in the outer-rise areaof the Sunda arc. This earthquake is known to be anomalous because of its extensive tsunami,which is disproportionate of its magnitude. The tsunami affected the coastal areas over 950 kmalong the Sumatran coast. We investigated this earthquake using the historical seismogramswe could collect from several seismological observatories. We examined the P-wave arrivaltimes listed in the Strassburg Bulletin (1912) and other station bulletins. The scatter of theObserved−Computed traveltime residuals ranges from –30 to 30 s, too large to locate the eventaccurately. The uncertainty of the epicentre estimated from an S-P grid-search relocation studyis at least 1◦ (∼110 km). We interpreted the Omori seismograms from Osaka, Mizusawa andTokyo, and the Wiechert seismograms from Gottingen and Uppsala by comparing them withthe seismograms simulated from modern broad-band seismograms of the 2002, 2008 and two2010 Sumatra earthquakes which occurred near the 1907 earthquake. From the amplitudeof Rayleigh waves recorded on the Omori seismograms we conclude that the magnitude ofthe 1907 earthquake at about 30 to 40 s is about 7.8 (i.e. 7.5 to 8.0). The SH waveformsrecorded on the Gottingen and Uppsala seismograms suggest that the 1907 earthquake is athrust earthquake at a shallow depth around 30 km. The most likely scenario is that the 1907earthquake initiated on the subduction interface, and slowly ruptured up-dip into the shallowsediments and caused the extensive tsunami. Although their quantity and quality are limited,historical seismograms provide key quantitative information about old events that cannot beobtained otherwise. This underscores the importance of preserving historical seismograms.
Key words Seismic cycle; Tsunamis; Earthquake source observations; Subduction zoneprocess; Dynamics and mechanics of faulting; Indian Ocean.
1 I N T RO D U C T I O N
Since the study of Newcomb & McCann (1987) seismic hazard inthe Sunda arc has attracted much attention of seismologists. Recentgeological studies of the history of coastal uplift and subsidenceextracted from corals (Natawidjaja et al. 2004; Sieh et al. 2008)combined with detailed studies of recent great earthquakes (theMw = 9.2 2004 Sumatra-Andaman earthquake (e.g. Lay et al. 2005),the Mw = 8.6 2005 Nias earthquake (e.g. Hsu et al. 2006; Koncaet al. 2007), and the Mw = 8.3 2007 Bengkula earthquake (e.g.Konca et al. 2008)) have clarified the rupture history along Sumatra
and contributed significantly to our understanding of seismic hazardin this region.
Among all the earthquakes in the region, the 1907 earthquake(January 4, 05:19, M = 7.6, 2◦N, 94 1/2◦E, depth = 50 km,Gutenberg & Richter (1954), Fig. 1) is known to be anomalous. De-spite its moderate magnitude, a report from the Royal Magnetic andMeteorological Observatory in Batavia (page 132–141, KoninklijkMagnetisch en Meteorologisch Observatorium te Batavia 1909),Visser (1922) and Newcomb & McCann (1987) noted that itstsunami affected the large areas extending nearly 950 km alongthe Sumatran coast. In fact, during the 2005 Nias earthquake, the
358 C© 2010 The Authors
Journal compilation C© 2010 RAS
Geophysical Journal International
The 1907 Sumatra earthquake 359
Figure 1. The location of the 1907 Sumatra earthquake and recent seismic-ity. (Adapted from Briggs et al. (2006), with inclusion of the 2007 sourcesfrom Konca et al. (2008). Courtesy of the Tectonic Observatory, CaliforniaInstitute of Technology).
experience with the 1907 earthquake apparently prompted the res-idents of the Simeulue Is. to take shelter on higher grounds imme-diately after they felt shaking, thereby saving many lives (McAdooet al. 2006). Gutenberg & Richter (1954) located it seaward of thetrench, near the outer-rise region (Fig. 1), which implies that it isnot a typical subduction thrust earthquake. Unfortunately, becauseof the lack of sufficient instrumental data, its seismological charac-teristics are poorly known. Should an earthquake of this type occursomewhere today, it may pose serious tsunami and shaking hazardsalong the coastal areas. If the 1907 earthquake is an outer-rise earth-quake, as is inferred from its location, it may cause much strongershaking along the coastal areas than ordinary thrust earthquakes.Ammon et al. (2008) demonstrated that short-period ground mo-tions with a period of 1 to 20 s from the 2007 large outer-riseearthquake in the Kuril Is. (Mw = 8.1) are four to seven times largerthan those of the 2006 megathrust earthquake (Mw = 8.3) whichoccurred two months earlier in the same region.
In view of the importance of this earthquake, we attempted tocollect as many historical seismograms as possible so that we canbetter understand its seismological characteristics. Unfortunately,at many seismological stations and observatories, old seismogramsare rapidly being lost, partly because of deterioration of the record-ing paper, lack of space, or lack of maintenance personnel andinfrastructure, and we could not find many records. Nevertheless,we could find a few good seismograms that proved extremely usefulfor unravelling some key characteristics of this unusual earthquake.
2 T S U NA M I
The anomalously large tsunami generated by this earthquake hasbeen described by a report from the Royal Magnetic and Me-
Figure 2. The areas affected by the tsunami caused by the 1907 Suma-tra earthquake (modified from Newcomb & McCann (1987)). The redstar shows the location given by Gutenberg & Richter (1954). Theblue star indicates the relocated epicentre (see Appendix A). The pur-ple star indicates the location of the 2002 (2.65◦N, 95.99◦E) and 2008(2.69◦N, 95.98◦E) Sumatra earthquakes (Mw = 7.2 and 7.3, respectively)used for waveform comparison. The dashed green line indicates the in-ferred initial rupture zone. The inset shows the mechanisms of the 2002and 2008 Sumatra earthquakes taken from the Global CMT solutions(http://www.globalcmt.org/CMTsearch.html).
teorological Observatory in Batavia (Koninklijk Magnetisch enMeteorologisch Observatorium te Batavia 1909) and Visser (1922).According to these authors, on January 4, 1907, destructive tidalwaves devastated the Isle of Simaloer (Simeulue), and were ob-served all along the coasts of Atjeh (Aceh), Tapanoeli and theMentawei Isles (Mentawai Is.). According to Newcomb & McCann(1987) (Fig. 2) this event caused tsunamis that devastated SimeulueIs. and extended over 950 km along the Sumatran coast. People onNias Is. were not able to stand. Islands on the seaward side of NiasIs. and towns on the seaward side of the Batu Island were devastatedby the sea wave. Monecke et al. (2008) note that a sand sheet oflimited extent found in the tsunami deposit at Aceh might corre-late with a tsunami of the 1907. Judging from these descriptions,it appears that the 1907 earthquake caused unusually widespreadtsunami for an earthquake with an estimated magnitude of about7.6. Unfortunately, we could not find any tide gauge data for thisevent.
3 L O C AT I O N
Gutenberg & Richter (1954) located this event at 2◦N, 94 1/2◦E(depth = 50 km) (Figs 1 and 2) seaward of the Sunda trench, nearthe outer-rise. At a face value, this location suggests a possibility thatit is a large outer-rise event. This location is very different from that(2.00◦N and 96.25◦E) reported by T. H. Staverman and published in
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
360 H. Kanamori, L. Rivera and W.H.K. Lee
Figure 3. Top: Copy of the Gutenberg’s notepad page for the 1907earthquake, and the relocation diagram (P. 696, Richter 1958). Bottom:Observed−Computed (O–C) traveltime residuals computed from the P ar-rival times reported in the Strassburg Bulletin.
Szirtes (1912). Staverman probably used S-P times. We reviewed theGutenberg’s notepad (here Gutenberg’s notepad refers to the origi-nal Gutenberg’s notes which contain the information on the locationand magnitude of the events listed in Gutenberg & Richter’s (1954)Seismicity of the Earth. All available notepads were microfilmed(Goodstein et al. 1980), and the original copies are archived at theCalifornia Institute of Technology). Fig. 3 illustrates Gutenberg’srelocation procedure. The scatter of the observed–computed travel-times (O–C residuals) is very large. Gutenberg probably applied aweighting scheme on the basis of his experience in his attempt tolocate this event, and he must have been aware of the large uncer-tainty in location. However, when the final location was entered inthe catalog of Seismicity of the Earth, the information on the size ofuncertainty was lost. Although it is difficult to accurately estimatethe uncertainty in the epicentre with the large scatter of O–C resid-uals (Fig. 3), our S-P grid-search relocation analysis suggests thatit is at least 1◦ (∼110 km) (see Appendix A). Although the locationis very uncertain, this information is useful because it allows us toaccommodate the possibility that this event could be of any typeother than an outer-rise event, such as a thrust event on the plateinterface.
We investigated the uncertainties in the location using S-P timesfrom 12 well-distributed stations. The use of S-P times has advan-tage because it can minimize the effect of clock errors that wereoften very large before the 1920s (until radio timing signals becamecommonly available). We used the JLoc interactive, direct-search,location software (Lee & Baker 2006; Lee & Dodge 2007) using amodern velocity model, AK135 by Kennett et al. (1995). A focaldepth is fixed at 20 km. The epicentre thus relocated is at 2.48◦Nand 96.11◦E, and is shown in Fig. 2. Because of the large error intraveltimes, the epicentre location estimated from S-P times has anerror of at least 1◦. More details are given in Appendix A.
4 WAV E F O R M S
We searched for old seismograms of this event by inquiring manyseismic stations in the world. As is described in Appendix C(Table C1), old seismograms have been lost at many stations inthe world, and despite our extensive efforts, we could find onlya few Omori seismograms from Japan, and Wiechert seismogramsfrom Europe that are useful for our analysis. Nevertheless, by exam-ining these seismograms, we could determine a few key attributesof this earthquake. In what follows, we describe the details of theseismograms we investigated.
4.1 Osaka (34.70◦N, 135.52◦E)
Fig. 4 shows the Omori seismogram recorded at Osaka, Japan.Although the original seismogram was discarded, a hand-copiedseismogram was published in the Report of Osaka Observatory (Re-port on Omori Horizontal Pendulum Seismograph Observations inOsaka for the two years 1906–1907 published by Osaka Observa-tory 1908). According to the Report, the Omori seismograph atOsaka had a free period, T0, of 27s, and the static magnification,V , of 20. Omori seismographs did not have any specific dampingdevice, and the actual damping was probably due to air friction, andsolid frictions at the pivot and the recording stylus. Occasionally, acalibration signal is given either in the beginning or at the end of therecord from which the damping ratio ε can be estimated (Richter1958, page 216). Unfortunately, no calibration signal is recorded onthis seismogram. Four distinct wave trains can be identified on theseismogram. As we will show later, they can be identified as P, S,‘Love’, and Rayleigh waves. ‘Love’ means that it is a combinationof multiple S phases and Love waves. In those days, the polarity ofthe seismogram was seldom written on the record, but this recordhas polarity clearly marked: upward motion on the record meanswestward ground motion.
To interpret this seismogram, we compared it with a simulatedOmori seismogram computed from a broad-band seismogram ofthe 2002 Sumatra earthquake (2002 November 2, 2.65◦N; 95.99◦E,23 km, Mw = 7.2) recorded at ABU (Abuyama, Japan, 34.864◦N,135.571◦E, 138 m). This earthquake is a thrust earthquake thatoccurred in the same area as the 1907 earthquake (Fig. 2). Thedistance between Osaka and ABU is 17 km.
The simulation was made by removing the instrument responsefrom the E-W component of the broad-band record, and convolvingit with the response of the Omori seismograph. As mentioned above,the effective damping of the Omori seismograph is not known. Thus,we vary the damping constant h from 0.05 to 0.4, compare thecomputed seismograms with the observed (Fig. 5), and chose h thatyields a seismogram most similar to the observed. The dampingratio ε is related to the damping constant h by ε = exp( πh√
1−h2).
We found that h = 0.2 is most appropriate, but inevitably someuncertainty is involved in this choice. As h decreases, the amplitudeof simulated seismogram increases. If h = 0.1 and 0.3, then theamplitude of the simulated seismogram becomes 1.31 and 0.83times, respectively, of that for h = 0.2. This range will be consideredin estimating the magnitudes. The simulated Omori seismogram forthe EW component record at Osaka is shown on Fig. 4. The overallwaveform of the observed record is similar to that of the simulatedrecord in terms of the relative amplitude of P, S, Love, and Rayleighwaves, which suggests that the fault geometry of the two events issimilar.
Fig. 6 shows the radiation patterns of P, SV , SH , and surfacewaves (both Love and Rayleigh waves) computed for the mechanism
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
The 1907 Sumatra earthquake 361
Figure 4. Comparison of the Omori seismogram (EW component) of the 1907 Sumatra earthquake recorded at Osaka, Japan, with a simulated Omoriseismogram computed from the seismogram of the 2002 Sumatra earthquake (Mw = 7.2) recorded at ABU (Abuyama), Japan. Here a damping constant, h, of0.2 is assumed.
Figure 5. The effect of damping constant, h, on the amplitude and waveform of simulated Omori seismograms computed from the broad-band record of the2002 Sumatra earthquake recorded at ABU (Abuyama, Japan). The damping constant, h, and the peak-to-peak amplitude of the wave at the free period aregiven on the right. For h < 0.1, the S group and Love wave become too continuous compared with the observed. To distinguish the waveforms for h = 0.2, 0.3,and 0.4 is difficult, but h much larger than 0.3 is unlikely for an instrument without any specific damping device. The peak-to-peak amplitudes of the wave atthe free period are 6.3, 4.8, and 4.0 for h = 0.1, 0.2, and 0.3, respectively.
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
362 H. Kanamori, L. Rivera and W.H.K. Lee
Figure 6. Top: The radiation patterns of P, SH, and SV phases for the 2002 Sumatra earthquake. The orthogonal nodal planes are shown for P wave, and thepositive and negative SH and SV waves are shown by blank and filled areas. The locations of the Japanese and European stations used in this paper are sketchedon the stereographic diagrams. Bottom: Schematic radiation patterns of Rayleigh and Love waves as a function of azimuth for the 2002 Sumatra earthquake.The amplitude is in an arbitrary unit; only the pattern is relevant here.
of the 2002 Sumatra earthquake. Japan is located in the directionclose to the nodes of SH and Love wave radiation patterns, butis close to the peak of Rayleigh-wave radiation pattern. Thus, theRayleigh-wave amplitude is robust with respect to small changes inthe mechanism.
The observed Rayleigh-wave amplitude is about 4 times largerthan the simulated, suggesting a surface-wave magnitude differenceof 0.6. However, because of the uncertainty in h, the magnitude esti-mate is correspondingly uncertain. We will estimate the magnitudelater using all the seismograms we investigated.
The P-wave first motion on the observed record appears to beeastward, which means an ‘up’ first-motion. Since the P-wave firstmotion in Japan from low-angle megathrust earthquakes in Sumatrais ‘up’ (Fig. 6), and that of outer-rise normal-fault earthquakesis most likely ‘down’, this observation suggests that this event isprobably a thrust event rather than a normal-fault event.
4.2 Mizusawa, Japan (39.13◦N, 141.13◦E)
Fig. 7 shows the Omori seismogram of the 1907 Sumatra earthquakerecorded at Mizusawa, Japan. The free period, T0, and the magni-fication, V , are 30 s and 20, respectively. We made a similar com-parison of this record with a simulated seismogram computed fromthe seismogram of the 2002 Sumatra earthquake recorded at the
station KSN (Kesen-numa, Japan, 38.976◦N, 141.530◦E, 260 m).The distance between Mizusawa and Kesen-numa is 39 km. In gen-eral, major features of the observed and the simulated records aresimilar, confirming the conclusion obtained from the Osaka record.The ratio of the observed to simulated peak-to-peak amplitudes ofthe Rayleigh wave is about 4.3, which translates to a magnitudedifference of 0.6.
Fig. 8 shows a similar comparison for the NS component. Thegain of the NS component is much lower, 9, than the EW componentfor unknown reason. A strange observation is that the Love wave isnot obvious on this component. Since the backazimuth at Mizusawais 240◦, Love wave should be stronger on the NS component regard-less of the mechanism. This observation remains unexplained, butS and Love waves are nodal at Mizusawa; we use only Rayleighwaves for the amplitude comparison.
4.3 Tokyo, Japan, (35.71◦N, 139.77◦E)
At Hongo in Tokyo, several Omori seismographs were in use (Omori1907). Fig. 9 is the one in Tokyo at the Seismological Institute ofTokyo Imperial University where Professor Omori was working.This is probably the same record as that Professor Omori investi-gated himself. The period, T0, and the magnification, V , are 41.5 sand 30, respectively.
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
The 1907 Sumatra earthquake 363
Figure 7. Comparison of the Omori seismogram (EW component) of the 1907 Sumatra earthquake recorded at Mizusawa, Japan, with a simulated Omoriseismogram computed from the seismogram of the 2002 Sumatra earthquake (Mw = 7.2) recorded at KSN (Kesen-numa, Japan).
Figure 8. Comparison of the Omori seismogram (NS component) of the 1907 Sumatra earthquake recorded at Mizusawa, Japan, with a simulated Omoriseismogram computed from the seismogram of the 2002 Sumatra earthquake (Mw = 7.2) recorded at KSN (Kesen-numa, Japan).
We computed a simulated record from a broad-band seismogramat TSK (Tsukuba, 36.214◦N, 140.090◦E, 174 m) of the 2008 Suma-tra earthquake (2008 February 20, 2.69◦N; 95.98◦E, 15 km, Mw =7.3) which is located close to the 2002 Sumatra earthquake (Fig. 2).
The 2002 Sumatra earthquake was not recorded at TSK. The dis-tance between Tokyo and Tsukuba is 63 km. The correspondencebetween the observed and simulated record is excellent. Since theRayleigh wave went off scale, we cannot measure the amplitude
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
364 H. Kanamori, L. Rivera and W.H.K. Lee
Figure 9. Comparison of the Omori seismogram (EW component) of the 1907 Sumatra earthquake recorded at Hongo, Tokyo Japan, with a simulated Omoriseismogram computed from the seismogram of the 2008 Sumatra earthquake (Mw = 7.3) recorded at TSK (Tsukuba, Japan).
ratio, but the observed to the simulated ratio is at least 4.9, suggest-ing that the surface-wave magnitude of the 1907 earthquake is atleast 0.7 units larger than that of the 2008 earthquake.
Fig. 10 shows an NS component Omori seismogram recorded atHongo. This is one of few NS component seismograms; most of theold Omori seismograms at Hongo now archived are EW compo-nent. On this record, a calibration signal is recorded at the bottomof the record from which we could estimate the damping ratio ε
(or damping constant h). Comparing the appearance of this seismo-gram with the one on Fig. 9, we can immediately see the differencein damping. The seismogram on Fig. 10 is close to a harmonicoscillation with the free period (40.0 s). The damping constant hestimated from the calibration signal is approximately 0.05. Al-though the waveforms of the observed and simulated records arevery similar, the amplitude measurement is not reliable because ofthe extremely small damping. A small error in h results in a largedifference in the amplitude of the simulated record. The ratio ofthe observed to the simulated Rayleigh-wave amplitudes is only 1.2with a corresponding magnitude difference of only 0.1.
4.4 Gottingen, Germany (51.550◦N, 9.967◦E), Uppsala,Sweden (59.850◦N, 17.635◦E), Potsdam, Germany(52.383◦N, 13.067◦E), and Hamburg, Germany (53.567◦N,9.983◦E)
Fig. 11 shows the two component Wiechert seismograms madeavailable to us by courtesy of the Gottingen Observatory. The po-larity of the Gottingen seismograms is clearly marked as N andE for downward motion on the record. The seismograph constants
are: NS T0 = 12.8s, ε = 4.2, V = 156; EW T0 = 13.3, ε = 5.6,V = 156. The most notable is the sharp S wave recorded on the NScomponent. Also shown on this figure are the two-component sim-ulated Wiechert seismograms computed from the seismograms ofthe 2002 Sumatra earthquake recorded at the station BFO, Germany(48.332◦N, 8.331◦E). Since the backazimuth at Gottingen from the2002 Sumatra earthquake is 98.5◦, the NS component is essentiallythe transverse component. The overall similarity between the 1907and 2002 seismograms concerning the energy partition between NS(SH component) and EW (SV component) and the large sharp SHwave on the NS component is evident, which suggests that the 1907and the 2002 events have a similar mechanism.
The copies of the Wiechert seismograms of the Uppsala Ob-servatory are faint, but the S wave on the NS component istraceable (Fig. 12). Unfortunately, the polarity is not marked onthe Uppsala seismograms. Different Wiechert seismographs canhave different polarity settings and all the possible combinations(upward = East, upward = West, upward = North, upward =South) are indeed found in the old Wiechert seismographs oper-ating early in the 20th century around the world (Charlier & VanGills 1953; M. Cara, personal communication, 2007). Accordingto Bungum et al. (2009), their test using the first-motion directionof the events with ‘known polarity’ and a direct test on the Up-psala seismograph indicated that upward motion on the NS andEW component records corresponds to southward and westwardground motion, respectively. We confirmed their result by check-ing the polarity of the P-wave first motion of the 1933 Sanriku,Japan, earthquake (Mw = 8.4). The studies of Matuzawa (1942)and Kanamori (1971) indicate that the P-wave first motion at
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
The 1907 Sumatra earthquake 365
Figure 10. Comparison of the Omori seismogram (NS component) of the 1907 Sumatra earthquake recorded at Hongo, Japan, with a simulated Omoriseismogram computed from the seismogram of the 2008 Sumatra earthquake (Mw = 7.3) recorded at TSK (Tsukuba, Japan). Here, a damping constant, h =0.05 is estimated from the calibration signal on the record.
Figure 11. Comparison of the NS and EW components of the 1907 Sumatra earthquake recorded at Gottingen, Germany, with simulated Wiechert seismogramscomputed from the seismogram of the 2002 Sumatra earthquake recorded at BFO (Black Forest Observatory, Germany). Note the overall similarity of theenergy partition between NS (SH) and EW (SV) components, and the large S wave on the NS component.
Uppsala for this earthquake must be ‘north’, and the recorded mo-tion on the NS component is down. Fig. 12 compares the NS com-ponent of the Gottingen seismogram and the Uppsala seismogram(T0 = 10s, ε = 5 (i.e. h = 0.46), and V = 182).
The seismogram we obtained from the Potsdam station is notmarked at all regarding the type of instrument, component, andpolarity. However, judging from the waveform, we assumed that itis the NS component (S is up on the record) Wiechert seismogram,
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
366 H. Kanamori, L. Rivera and W.H.K. Lee
Figure 12. Comparison of the S wave on the NS component seismograms at Gottingen and Uppsala. Note the similarity of the waveform, and the northwardfirst motion. The Potsdam seismogram is shown just for comparison.
and include it on Fig. 12 in comparison with the records fromGottingen and Uppsala. This record is not used for analysis butis shown only for comparison purposes. The first few cycles ofS wave are similar between Gottingen and Uppsala. The Potsdamrecord is also similar suggesting that our assumption is probablycorrect.
Fig. 13 compares the NS component of S waveform recorded onthe Gottingen seismogram with the simulated Wiechert seismogramcomputed from the record of the 2002 Sumatra earthquake recordedat BFO. As shown in Fig. 12 the first motion of S wave is northwardat both Gottingen and Uppsala, and is the same as that of the 2002Sumatra earthquake; this suggests that the 1907 earthquake is athrust event similar to the 2002 event. This supports the somewhatweaker conclusion we made earlier from the first motion recordedat Osaka.
The waveform similarity deteriorates for the second pulse. Thiscould be due to the difference in depth, source pulse shape, etc. Toillustrate the situation, we show on the lower-left corner of Fig. 13the moment-rate function of the 2002 event obtained by inversionof 64 P waveforms. The half duration is about 15s. The difference inthe frequency content between the 1907 and 2002 events suggeststhat the moment-rate function for the 1907 event can be slightlybroader. A synthetic waveform computed for a point source at adepth of 30 km having a triangular moment-rate function with ahalf duration of 20 s can match the polarity of the second pulse asshown in Fig. 13. Thus, the waveform mismatch can be partially dueto a small difference in the source moment rate function betweenthe 1907 and 2002 events.
Fig. 14 shows the S waveforms simulated from the seismo-grams of four recent Sumatra earthquakes recorded at BFO (2002
November 2, Mw = 7.2, d = 23 km, 2008 February 2, Mw = 7.3,d = 14.9 km, 2010 April 6, 2.05◦N, 96.71◦E, Mw = 7.8, d = 19.7km, and 2010 May 9, 3.38◦N, 95.79◦E, Mw = 7.2, d = 38 km).These events are located close to each other with the depths from14 to 38 km (GCMT: http://www.globalcmt.org), probably on thesubduction-zone plate interface (Sladen et al. 2009; Tilmann et al.2010). Although the displacement records of these events are rel-atively simple and similar to each other, the simulated Wiechertrecords are complex and cannot be related to the depth in a simpleway. Thus, it is difficult to estimate the depth of the 1907 eventaccurately from the waveform, but the waveform complexity of the1907 event suggests that this event is comparable in depth to these4 Sumatra earthquakes, and is located on the plate boundary inter-face. We use 30 km as a nominal depth of the 1907 earthquake, butthe implied uncertainty is from 14 to 38 km, the range of the fourSumatra earthquakes.
The amplitude ratio of the SH wave of the 1907 to the 2002 eventis approximately 2.0 and 1.8 at Gottingen and Uppsala, respectively(the simulated Wiechert seismogram for Uppsala is not shown, butis similar to that for Gottingen).
Another interesting observation is that while the waveforms dur-ing the first 100 s are similar, the record of the 1907 event exhibitssignificant amounts of energy for another 100 s (Figs 11 and 12),during which no significant energy is seen on the record of the2002 event. This suggests that the 1907 earthquake had an extendedrupture following the initial rupture.
The Hamburg Wiechert seismograms show S waves and surfacewaves, but the traces are tangled up and hard to trace and timingmarks are difficult to read. We used these records only for magnitudedetermination.
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
The 1907 Sumatra earthquake 367
Figure 13. Comparison of the S waveform recorded at Gottingen (top) and a simulated Wiechert seismogram computed from the seismogram of the 2002Sumatra earthquake recorded at BFO. Note that the first motion is northward for both the 1907 and 2002 events, suggesting that the 1907 earthquake had athrust mechanism. The waveform agreement deteriorates in the second pulse. This could be due to the difference in depth, source pulse shape, etc. The momentrate function for the 2002 earthquake shown on the lower-left corner has a half duration of about 15 s. The difference in the frequency content between the1907 and 2002 events suggests that the moment-rate function for the 1907 event can be slightly broader. The bottom trace is a synthetic waveform computedfor a point source with a half duration of 20 s at a depth of 30 km.
Figure 14. Comparison of S waveforms (NS component) at BFO for four Sumatra earthquakes at depths from 14 to 38 km. Top: Displacement records.Bottom: Simulated Wiechert seismograms computed with the instrument constants of the NS component of the Gottingen Wiechert seismograph.
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
368 H. Kanamori, L. Rivera and W.H.K. Lee
Table 1. (a) Magnitude estimations from Omori seismograms; (b) Magnitude estimationsfrom Wiechert Seismograms; (c) Estimation of mB from Wiechert seismograms.
(a)Station Component Amplitude ratio �MS MS
Osaka EW 3.2 to 5.1 0.5 to 0.71 7.65 to 7.86Mizusawa EW 4.3 to 6.7 0.63 to 0.83 7.78 to 7.98Mizusawa NS 1.4 to 2.1 0.15 to 0.32 7.30 to 7.47Hongo EW >5.4 to 8.5 >0.73 to 0.93 >7.96 to 8.16Hongo NS 1.24 0.09 7.14
(b)Station Component Amplitude (μm) Period (s) MS
Uppsala NS 350 20 7.56EW 171 20
Hamburg NS 700 20 7.6EW 400
(c)Station Component Amplitude (μm) Period (s) mB
Uppsala NS 90.7 17 7.3Gottingen NS 191 26 7.7
5 S U M M A RY O F M A G N I T U D EE S T I M AT E S
Table 1 summarizes the results of amplitude measurements. Weestimated the magnitude difference from the amplitude ratios ofthe observed to simulated records, and estimated the surface-wavemagnitude, MS , using the reference values for the 2002 (MS = 7.15)and 2008 (MS = 7.25) Sumatra earthquakes (Appendix B). Sincethe damping of the NS component of the Hongo record is verysmall, we consider the MS estimate from this record unreliable. Dis-regarding this estimate and considering the lower bound estimatedfrom the EW component at Hongo, we prefer a range 7.5 < MS <
8.0 (Table 1a).The amplitude measurements from the Wiechert seismograms at
Uppsala and Hamburg are summarized in Table 1b. The Uppsalaseismograms came with a detailed calibration sheet, and the am-plitude measurements are considered reliable. The Hamburg seis-mograms did not come with calibration, and we assume a gain, 20,at a period of 20 s, which is the value for this type of Wiechertseismograph. For these determinations, the traditional Gutenberg’s(1945) formula, MS = log(AH ) + 1.656 log � + 1.818 is used.The MS values from these records agree well with that listed inGutenberg & Richter (1954). However, as shown in Figs 6 and A1,the European stations are in the nodal azimuth of Rayleigh waves.If we correct for this effect from Figure A1, MS = 7.8 is probablymore appropriate.
From the amplitude of the SH wave recorded at Uppsala andGottingen (Table 1c), we obtain mB = 7.3, and 7.7, respectively.The mB value at Gottingen was obtained at a longer period than atUppsala. These values are comparable to what are expected of anordinary MS = 7.8 earthquake (Gutenberg 1956).
In summary, combining all these results, we estimate MS = 7.8,with an uncertainty range of +−0.25. From the mB and MS re-lation, the 1907 event does not seem to be particularly anomalousregarding the source radiation spectrum up to a period of 30 s. How-ever, there is some indication that the Omori EW seismogram atHongo that has a longer free period (40 s) than the others gavea larger MS (>7.96). The body-wave magnitudes at Uppsala andGottingen display a similar trend. These observations together could
be taken as evidence for the 1907 earthquake having an enhancedexcitation at longer periods.
6 T H E RU P T U R E P RO C E S S
Although the quality of old data is limited, we can synthesize theanalyses presented above and the description of macroseismic databy Visser (1922) and Newcomb & McCann (1987) as follows.
The location determined from teleseismic traveltimes is uncer-tain, and the uncertainty of the epicentre estimated from an S-Pgrid-search relocation study (Appendix A) is at least 1◦ (∼110 km).The comparison of the Wiechert seismogram recorded at Gottingenand Uppsala and that simulated from the record of the 2002 Suma-tra earthquake suggests that the initial rupture of the 1907 eventis a thrust event at a depth of approximately 30 km, and it wasfollowed by an extended rupture for at least 100s. From this obser-vation alone we cannot determine where the extended rupture was,but referring to Newcomb and McCann (1987) it is most likely sea-ward of the Nias and Simeulue Islands. According to Newcomb &McCann (1987), the islands on the seaward side of Nias and towns onthe seaward side of Batu Island were devastated by the sea wave.The green dashed line in Fig. 2 schematically shows the approxi-mate initial rupture zone thus inferred. From our seismic data withthe frequency band up to 40 s, we cannot determine the total size(i.e. Mw) of this event, but the magnitude determined at a period ofup to 40 s is probably between 7.5 and 8. The extent of the tsunamidamage (950 km) suggests that the Mw can be significantly largerthan this, but without tide gauge data we cannot estimate it.
7 D I S C U S S I O N
The 1907 Sumatra earthquake has all the attributes of tsunami earth-quakes (Kanamori 1972) and is similar to the 1896 Sanriku, Japan,earthquake and the 1946 Unimak Is., the Aleutians, earthquake.According to Visser (1922) and Newcomb & McCann (1987), the1907 earthquake caused strong shaking on the Nias and Simeu-lue Islands, but the intensities on Sumatra were only moderate.Although no strong shaking was reported for most other tsunami
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
The 1907 Sumatra earthquake 369
Figure 15. Coseismic and postseismic slip models for the 2005 (March 28) Nias earthquake (Mw = 8.6). (after Hsu et al. 2006; Courtesy of Dr. Yaru Hsu)
earthquakes including the 1896 Sanriku, the 1946 Unimak Is., the1992 Nicaragua earthquake, the 1994 Java earthquake, and the 2006Java earthquake, these subduction zones do not have islands in theforearc region over the shallow portion of the megathrust. Thus,despite the difference in the report of strong shaking, the 1907earthquake may be similar to other tsunami earthquakes.
It is interesting to compare the 1907 event with the March 28,2005, Nias earthquake. Although the Nias earthquake is a greatearthquake with Mw = 8.6, no anomalously large tsunami wasgenerated. Hsu et al. (2006) made a detailed study of this earthquakeusing seismic and static (GPS, INSAR, Coral) data and obtained theresult shown in Fig. 15. The coseismic slip extended over nearly 300km right underneath the Nias-Smeulue island chain, and significantafterslip occurred on the plate boundary between the Nias Islandand the trench. However, no large coseismic slip occurred in theup-dip part of the plate boundary. In contrast, the large tsunamiassociated with the 1907 earthquake is most likely caused by a slipin the up-dip part of the subduction boundary. It is possible thatat the time of the Nias earthquake in 2005, not enough strain hadaccumulated there since 1907 to cause extensive tsunami. Briggset al. (2006) also note this possibility.
Earthquakes like the 1907 earthquake are less likely to leave dis-tinct uplift and subsidence records in corals because the primarydeformation was in a relatively narrow zone along the trench, beingfar from the coast. However, in assessing the seismic and tsunamihazard, it would be important to include the events like the 1907earthquake that pose extremely serious tsunami hazard to the pop-ulations in the coastal area.
The 1907 earthquake is somewhat similar to the 1605 Keichoearthquake (called ‘Keicho’ because it occurred in the Keichoera in the Japanese calendar) along the Nankai trough. Along theNankai trough, a series of great earthquakes have repeatedly oc-curred with an average interval of 120 years (Imamura 1928; Ando1975; Ishibashi 1981). However, the 1605 event is known to beanomalous, in that despite the limited extent of shaking, it gener-ated extensive tsunamis along the Japanese Islands. If the 1907 event
released the strain which had accumulated on a shallow boundaryover a long period of time, much longer than the average repeattime of regular large megathrust events like the 2005 Nias and2007 Bengkulu earthquakes, the rupture sequence of the 1907 typeearthquakes can be very different from that of the 2005 Nias typeearthquakes even if they occur in the same region. Without knowl-edge on the detailed rupture process of old earthquakes, it would bedifficult to separate these two groups of earthquakes and both typesof events tend to be treated together to estimate the average repeattime. The 1605 earthquake may have to be considered separatelyfrom the rest of more ‘regular’ megathrust earthquakes.
This behaviour raises another important question regardingtsunami potential of subduction zones where no large historicalevent has been documented (e.g, a segment of subduction zonesouth of Sanriku in Japan). Because most seismic and tsunami haz-ard mitigation measures heavily rely on the past experience, such‘quiet’ subduction zones tend to receive less attention, but slow ac-cumulation of strain in such subduction zones can lead to extremelyserious, though infrequent, tsunami hazard, and special attentionneeds to be paid to such possibilities.
8 C O N C LU S I O N
We have shown that even if old seismograms are few with poorlydocumented characteristics, we can use them to clarify certain as-pects of important historical earthquakes.
From the waveforms and the amplitude of Omori and Wiechertseismograms, we could determine that the 1907 Sumatra earthquakewas probably a thrust earthquake on the plate boundary, not an outer-rise earthquake as inferred from the location given by Gutenberg& Richter (1954). The magnitude is still uncertain but up to aperiod of 40 s, it is probably 7.8 (meaning between 7.5 and 8) withsome indication of increasing size with period from 10 to 40 s. Wenote, however, because of the limited number and quality of oldseismograms, our conclusion is subject to considerable uncertaintyand should be taken with caution. We tried to present as much raw
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
370 H. Kanamori, L. Rivera and W.H.K. Lee
data as possible to allow other investigators to examine them if someaspects of the conclusions are questioned.
If more seismograms with well-documented instrument charac-teristics are available, we would be able to harden our conclusion.The current situation (Lee & Benson 2008), however, is alarmingbecause old seismograms are being discarded at many seismolog-ical observatories and institutions because of the space, budgetaryand personnel limitations. We hope that this study has demonstratedthe value of old seismograms for unravelling key characteristics anddiversity of subduction-zone earthquakes which can be understoodwell only from data over an extended period of time. A better under-standing of the diversity of subduction-zone earthquakes is criticallyimportant for implementing comprehensive hazard mitigation mea-sures.
A C K N OW L E D G M E N T
We thank our colleagues at many seismological institutions andobservatories who helped us explore the availability and collectvaluable historical seismograms. Just to mention a few: ConnyHolmqvist, Kazuko Noguchi, Kenshiro Tsumura, Nobuo Hamada,Norihito Umino, Satoshi Hirahara, Torsten Dahm, Martin Leven,Thomas Jahr, Dieter Kurrle, Zhu Yuanqing, Thierry Camelbeeck,Suhardjono, Randell Teodoro, Antoniuo Pazos, Bernard Dost, BrianFerris, and Zurab Javakhishvili. Michel Cara helped us understandthe polarity convention of Wiechert seismographs. We have ben-efited a great deal from discussions with William McCann whoalso provided us with the base map for Fig. 2. We thank KerrySieh, William McCann and Johannes Schweitzer for providing uswith helpful comments during the final revision process. We thankDoug Dodge for coding the JLoc grid-search earthquake locationprogram. The copies of the pages of the Gutenberg notepad usedin this study were provided by the Archives of the California Insti-tute of Technology. We thank the IRIS Data Management Centerfor providing us access to the global broad-band sesimograms. Wethank the National Research Institute for Earth Science and Disas-ter Prevention (NIED) Japan for providing us access to the F-netbroad-band seismograms. Archiving and scanning project of Mizu-sawa seismograms was supported by Grant (PR2002) ‘The Researchon the Tonankai and Nankaido earthquakes’ from the ministry ofEducation, Culture, Science and Technology of Japan.
R E F E R E N C E S
Abe, K., 1981. Magnitudes of large shallow earthquakes from 1904 to 1980,Phys. Earth planet. Inter., 27, 72–92.
Ammon, C.J., Kanamori, H. & Lay, T., 2008. A great earthquake doubletand seismic stress transfer cycle in the central Kuril islands, Nature, 451,561–565.
Ando, M., 1975. Source mechanisms and tectonic significance of histor-ical earthquakes along the Nankai trough, Japan, Tectonophysics, 27,119–140.
Briggs, R.W. et al., 2006. Deformation and slip along the Sunda Megathrustin the great 2005 Nias-Simeulue earthquake, Science, 311, 1897–1901.
Bungum, H., Pettenati, F., Schweitzer, J., Sirovich, L. & Faleide, J.I.,2009. The 23 October 1904 MS 5.4 Oslofjord earthquake: Reanalysisbased on macroseismic and instrumental data, Bull. seism. Soc. Am., 99,2836–2854.
Charlier, C. & Van Gils, J.M., 1953. Liste des Stations Seismologi-ques Mondiales, Available at: http://www.iris.edu/seismo/info/stations/Charlier1953.pdf] in 2 volumes, no page number, Observatoire Royal deBelgique, Uccle.
Crotwell, H.P., Owens, T.J. & Ritsema, J., 1999. The TauP Toolkit: flexibletraveltime and ray path utilities, Seism. Res. Lett., 70(2), 154–160.
Geiger, L.C., 1912. Probability method for the determination of earthquakeepicenters from the arrival time only, Bull. St. Louis Univ., 8, 60–71.
Geller, R.J. & Kanamori, H., 1977. Magnitudes of great shallow earthquakesfrom 1904 to 1952, Bull. seism. Soc. Am., 67, 587–598.
Goodstein, J.R., Kanamori, H. & Lee, W.H.K., 1980. Seismology mi-crofiche publications from the Caltech archives, Bull. seism. Soc. Am.,70, 657–658.
Gutenberg, B., 1945. Amplitudes of surface waves and magnitudes of shal-low earthquakes, Bull. seism. Soc. Am., 35, 3–12.
Gutenberg, B. & Richter, C.F., 1954. Seismicity of the Earth and AssociatedPhenomena, 2nd edn, p. 310, Princeton Univ. Press, Princeton.
Gutenberg, B. & Richter, C.F., 1956. Magnitude and energy of earthquakes,Ann. Geofis., (Rome), 9, 1–15.
Hsu, Y.-J. et al., 2006. Frictional afterslip following the 2005 Nias-Simeulueearthquake, Sumatra, Science, 312, 1921–1926.
Imamura, A., 1928. On the seismic activity of central Japan, Jpn. J. astr.Geophys., 6, 119–137.
Ishibashi, K., 1981. Specification of a soon-to-occur seismic faulting inthe Tokai District, central Japan, based on seismotectonics. in Earth-quake Prediction: An International Review Maurice Ewing Ser., Vol. 4,pp. 297–332, eds Simpson, D.W. & Richards, P.G., American GeophysicalUnion, Washington, DC.
Jeffreys, H. & Bullen, K.E., 1940. Seismological Tables. British Associationfor the Advancement of Science, London.
Kanamori, H., 1971. Seismological evidence for a lithospheric normalfaulting–the Sanriku earthquake of 1933, Phys. Earth planet. Inter., 4,289–300.
Kanamori, H., 1972. Mechanism of Tsunami earthquakes, Phys. Earthplanet. Inter., 6, 346–359.
Kennett, B.L.N. & Engdahl, E.R., 1991. Traveltimes for global earthquakelocation and phase identification, Geophys. J. Int., 105, 429–465.
Kennett, B.L.N., Engdahl, E.R. & Buland, R., 1995. Constraints on seis-mic velocities in the Earth from traveltimes, Geophys. J. Int., 122, 108–124.
Konca, A.O. et al., 2008. Partial rupture of a locked patch of the Sumatramegathrust during the 2007 earthquake sequence, Nature, 456, 631–635,doi:610.1038/nature07572.
Konca, A.O. et al., 2007. Rupture kinematics of the 2005 Mw 8.6 Nias-Simeulue earthquake from the joint inversion of seismic and geodeticdata, Bull. seism. Soc. Am., 97, S307–S322.
Kozak, J., 2001. 100-year anniversary of the first international seismologicalconference, Studia geoph. et geod., 45, 200–209.
Lay, T. et al., 2005. The great Sumatra-Andaman earthquake of 26 December2004, Science, 308, 1127–1133.
Lee, W.H.K. & Baker, L.M., 2006. Development of a direct search softwarepackage for locating poorly constrained earthquakes (abstract), Seism.Res. Lett., 77, 291–292.
Lee, W.H.K. & Benson, B.R., 2008. Making non-digital-recorded seismo-grams accessible online for studying earthquakes, in Historical Seis-mology, pp. 403–427, eds J. Frechet, J., Meghraoui, M. & Stucchi, M.,Springer, Berlin.
Lee, W.H.K. & Dodge, D.A., 2007. Development of a direct search softwarepackage for poorly constrained earthquakes, in Seismology Technical Re-port, pp. 6–27, Central Weather Bureau, Taiwan.
Matuzawa, T., 1942. Seismometrische Untersuchungen des Erdbebens vom2. Marz 1933. : IV. Raumverteilung der Wellenstrahlung aus dem Herd,in Bulletin of Earthquake Research Institute, Vol. 20, pp. 162–171, TokyoUniversity.
McAdoo, B.G., Dengler, L., Prasetya, G. & Titov, V., 2006. How an oral his-tory saved thousands on Indonesia’s Simeulue Island during the December2004 and March 2005 tsunamis, Earthq. Spectra, 22, S661–S669.
Monecke, K., Finger, W., Klarer, D., Kongko, W., McAdoo, B.G. & Moore,A.L., 2008. A 1,000-year sediment record of tsunami recurrence in north-ern Sumatra, Nature, 455, 1232–1234.
Natawidjaja, D.H., Sieh, K., Ward, S.N., Cheng, H., Edwards, R.L.,Galetzka, J. & Suwargadi, B.W., 2004. Paleogeodetic records of seismic
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
The 1907 Sumatra earthquake 371
and aseismic subduction from central Sumatran microatolls, Indonesia, J.geophys. Res., 109, B04306, doi:10.1029/2003JB002398.
Newcomb, K.R. & McCann, W.R., 1987. Seismic history and seismotecton-ics of the Sunda arc, J. geophys. Res., 92, 421–439.
Press, W.H., Flannery, V, Teukolsky, S.A. & Vetterling, W.T., 1986. Numer-ical Recipes: The Art of Scientific Computing, Cambridge Univ. Press,Cambridge.
Omori, F., 1907. Notes on the Valparaiso and Aleutian earthquakes of August17, 1906, Bulletin of the Imperial Earthquake Investigating Committee,vol. 1, pp. 75–113.
Osaka Observatory, 1908. Report on Omori Horizontal Pendulum Seismo-graph Observations in Osaka for the Two Years 1906–1907.
Richter, C.F., 1958. Elementary Seismology, pp. 768, W. H. Freeman, SanFrancisco.
Royal Magnetic and Meteorological Observatory, Batavia, 1909. Vulka-nische verschijnselen en aardbevingen in den Oost-Indischen Archipelwaargenomen gedurende het jaar 1907, in Natuurkundig Tijdschrift voorNederlandsch-Indie, pp. 117–201, ed. Bemmelen, W.V., Nederlandsch-Indie utigeven door de Koninklijke Natuurkundige Vereeniging in Nederl-Indie, Welevreden.
Sieh, K. et al., 2008. Earthquake supercycles inferred from sea-level changesrecorded in the corals of west Sumatra, Science, 322, 1674–1678.
Sladen, A., Avouac, J.-P., Meltzner, A.J., Genrich, J.F., Galetzka, J.E., Sieh,K.E. & Natawidjaja, D.H., 2009. Source model of the 2008, Mw = 7.4,Simeulue, Indonesia earthquake (abstract), T23B-1926, AGU Fall Meet-ing, 2009.
Szirtes, S., 1912. Registrierungen der besser ausgepragten seismischenStorungen des Jahres 1907, Erganzung zum seismischen Katalog.Veroffentlichungen des Zentralbureaus der Internationalen Seismologis-chen Assoziation. Serie B, Kataloge, Strassburg, pp. 1, 5–7.
Tilmann, F.J., Craig, T.J., Grevemeyer, I., Suwargadi, B., Kopp, H. & Flueh,E., 2010. The updip seismic/aseismic transition of the Sumatra megath-rust illuminated by aftershocks of the 2004 Aceh-Andaman and 2005Nias events, Geophys. J. Int., 181, 1261–1274, doi:1210.1111/j.1365-1246X.2010.04597.x.
Visser, S.W., 1922. Inland and Submarine Epicentra of Sumatra and Javaearthquakes, Koninklijk Magnetisch en Meteorologisch Observatorium teBatavia, 9, 1–14.
A P P E N D I X A : N O T E O N T H EE P I C E N T R A L L O C AT I O N
Szirtes (1912) provides the most complete list of arrival times for the1907 earthquake, and an epicentre location at 2.00◦N and 96.25◦Ebased on a report by T. H. Staverman. This epicentre differs sig-nificantly from the epicentre location published by Gutenberg &Richter (1954) at 2.00◦N and 94.50◦E. Just to get some ideas aboutthe uncertainty in location, we used the JLoc interactive, direct-search, location software (Lee & Baker 2006; Lee & Dodge 2007),and relocated the hypocentre. The JLoc program allows the user tochoose several different velocity models (including the Jeffreys &Bullen’s (1940) and several modern ones, such as iasp91 by Kennett& Engdahl 1991, and ak135 by Kennett et al. 1995), and is capableof computing station residuals (using the TauP traveltime calcula-tor by Crotwell et al. 1999), and their rms values from a specifiedhypocentre, thereby performing a grid-search on a user specified 3-D space grid for a “best fit” solution. JLoc also includes a simplexalgorithm (Press et al. 1986) to ‘home’ in for a final solution.
Because of large uncertainty in the station clock before radiotime signals became available in the 1920s, locating an earthquakeusing P- and/or S-arrival times using the standard Geiger (1912)method is difficult. The use of S-P and SKS-P traveltimes avoidsthe station clock problem, but the location accuracy still depends onhow accurately the P and S (or SKS) phases were ‘picked’, which in
Table A1. Observed traveltime for S-P or SKS-P from 12 well-distributedselected stations.a
Station Delta Azimuth Observedname (degree) (degree) Phase Traveltime (s) Code
Simla 34.09 330.0 S-P 336. SMIZi-ka-wei 37.52 36.8 S-P 351. ZKWPerth 38.53 152.8 S-P 378. PEROsaka 48.89 43.8 S-P 421. OSAIrkutsk 50.51 6.4 S-P 450. IRKTiflis 60.67 317.9 S-P 504. TIFPulkova 76.25 331.7 S-P 580. PULCapetown 80.97 235.3 S-P 612. CTOMessina 81.40 308.2 S-P 607. MESUppsala 82.46 330.1 S-P 613. UPPGottingen 86.04 321.4 S-P 628. GTTHonolulu 103.82 67.5 SKS-P 624. HON
Note: aDelta is the epicentral distance (in degrees) from 2◦ 00’ N and 96◦15’E, and Azimuth is from the epicentre to the station measured from theNorth in degrees.
Table A2. Comparison of epicentre locations by different authors (rmsvalue is computed using the residuals of the same set of stations given inTable A1, and a modern velocity model, AK135 by Kennett et al. 1995). Afocal depth of 20 km is assumed for all solutions.
Author (reference) Latitude Longitude rms (sec)
Staverman (Szirtes 1912) 2.00N 96.25E 8.15Gutenberg & Richter (1954) 2.00N 94.50E 9.41This study 2.48N 96.11E 7.80
turn depends on the clarity of the seismic phases and the time markresolution of the seismograms. For a S-P (and SKS-P) relocation,we selected 12 well-distributed stations from a total of 42 availablestations. We added arrival times of Uppsala, Honolulu, and Sitkafrom their station bulletins to the compilation of Szirtes (1912).These selected data are shown in Table A1. For comparison, therms values computed for the Gutenberg & Richter’s (1954) locationand the Staverman’s location are shown in Table A2.
We conclude that the available arrival time data in1907 are sopoor (probable picking error of about 10 s for Milne seismogramsor about 3 s for Omori or Wiechert seismograms; in addition, therewere large clock errors, exceeding tens of seconds at some stations)that epicentre location has an error of at least 1◦. The epicentre thusrelocated is at (2.48◦N, 96.11◦E). The 95 per cent confidence ellipseparameters are: Semi-major axis length = 290 km; Semi-minor axislength = 223 km, and strike of semi-major axis = 69◦.
A P P E N D I X B : M A G N I T U D EE S T I M AT I O N
The 2002 Sumatra and 2008 Sumatra earthquakes are given MS =7.6 and 7.5, respectively, in the USGS report. Since, the currentpractice of MS determination is slightly different from the practiceused by Gutenberg & Richter (1954), and since the number andazimuthal coverage of stations used in these determinations areunknown, we estimated MS using all the broad-band stations asfollows. We bandpass filter the vertical component of displacementover a narrow frequency band between 0.04 to 0.06 Hz, and measurethe peak amplitude, Az. Then, we use the IASPEI formula
MS = log(Az/T ) + 1.66 log � + 3.3
with T = 20 s. This formula is supposed to be used for the amplitudemeasured from the horizontal component, but Abe (1981) found
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
372 H. Kanamori, L. Rivera and W.H.K. Lee
Figure B1. Azimuthal distribution of MS of the 2002 and 2008 Sumatraearthquakes determined from bandpassed records.
no systematic difference in MS whether the horizontal or verticalcomponents are used. Then, to conform to the original definitionof Gutenberg (1945), we subtract 0.18 following the suggestions byGeller & Kanamori (1977) and Abe (1981).
The results are shown in Fig. B1. For the 2008 event for whichmany stations are available, the effect of radiation pattern is evi-dent. A similar, but incomplete, trend is seen for the 2002 event.The average MS is 7.15 and 7.23 for the 2002 and 2008 events,respectively. We use these values as reference.
A P P E N D I X C : AVA I L A B I L I T YO F S E I S M O G R A M S
We have contacted the observatories and institutions at the follow-ing locations (some are old names): Japan Meteorological Agency,Mizusawa Observatory, Earthquake Research Institute (Tokyo Uni-versity), Hamburg, Gottingen, Jena, Stuttgart, Strasbourg, Uppsala,Zikawei, Uccle, Batavia, Manila, San Fernando, Ebro, De Bilt,Christchurch, and Tbilisi. For the 1907 earthquakes, we could ob-tain records only from Osaka Observatory (hand-drafted copy),Mizusawa (Omori), Earthquake Research Institute (Omori),Gottingen (Wiechert), Potsdam 9probably Wiechert), Hamburg(Wiechert), Uppsala(Wiechert), Strasbourg (microfilm Wiechert),and San Fernando (Milne). The stations and institutions we con-tacted in this study are listed in Table C1.
Figure C1. (a) Professos Fusakichi Omori and Emil Wiechert at the oc-casion of the Internationale Seismologische Conferenz held in Strassburg,April 11–13, 1901 (Kozak (2001), with permission). (b) Omori seismograph(horizontal) and Wiechert seismograph (horizontal).
Considering the historical importance of the old seismograms,we include a photograph of Professors Wiechert and Omori sittingside by side at a rare occasion, and the seismographs to which theirnames are attached in Fig. C1.
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
The 1907 Sumatra earthquake 373
Tab
leC
1.L
isto
fst
atio
nsan
din
stit
utio
nsco
ntac
ted
inth
isst
udy.
Con
tact
info
rmat
ion
and
the
stat
usof
arch
ive
are
also
incl
uded
.
Sta
tion
1907
Sum
atra
reco
rdIn
stit
utio
nN
ame
–e-
mai
l–U
RL
Not
e
Osa
ka,J
apan
Om
ori(
copy
)
Miz
usaw
a,Ja
pan
Om
ori(
EW
,NS
)M
izus
awa
Lat
itud
eO
bser
vato
ryIn
good
cond
itio
n
Toky
o(H
ongo
),Ja
pan
Ear
thqu
ake
Res
earc
hIn
stit
ute
Dr.
Ken
jiS
atak
esa
take
@er
i.uto
kyo.
ac.jp
ww
w.e
ri.u
-tok
yo.a
c.jp
Man
yO
mor
isei
smog
ram
sin
good
cond
itio
n
Ham
burg
,Ger
man
yW
iech
ert(
H)
Inst
itut
fur
Geo
phys
ik,U
nive
rsit
atH
ambu
rgD
r.To
rste
nD
ahm
http
://w
ww
.geo
phys
ics.
zmaw
.de
Ver
yco
mpl
ete
coll
ecti
onW
iech
ert:
1905
–194
2,19
52–1
974
Hec
ker:
1906
–192
8
Got
ting
en,G
erm
any
Wie
cher
t(H
)In
stit
utfu
rG
eoph
ysik
,Uni
vers
itat
Got
ting
enw
ww
.geo
.phy
sik.
uni-
goet
ting
en.d
eV
ery
com
plet
eco
llec
tion
Wie
cher
t:01
-190
3to
06-2
005
Stu
ttga
rt,G
erm
any
No
reco
rdIn
stit
utfu
rG
eoph
ysik
,Uni
vers
itat
Stu
ttga
rtD
r.R
udol
fW
idm
er-S
chni
drig
wid
mer
@ge
ophy
s.un
i-st
uttg
art.d
eFa
irly
com
plet
eco
llec
tion
Om
ori-
Bos
chsi
nce
1905
Wie
cher
tsin
ce19
37
Jena
,Ger
man
yN
ore
cord
Inst
itut
fur
Geo
wis
sens
chaf
ten,
Uni
vers
itat
Jena
ww
w.g
eo.u
ni-j
ena.
deR
ecor
dsfr
omth
eJe
naan
dPo
tsda
mob
serv
ator
ies;
see
ww
w.g
eo.u
ni-j
ena.
de/g
eoph
ysik
/hom
e/ar
chiv
.htm
l
Str
asbo
urg,
Fran
ceW
iech
ert(
H)
Eco
leet
Obs
erva
toir
ede
Sci
ence
sde
laTe
rre
Dr.
Lui
sR
iver
alu
is.r
iver
a@un
istr
a.fr
eost
.u-s
tras
bg.f
r
Mic
rofi
lm(o
ften
poor
):W
iech
ert(
1904
–193
0),
Gal
itzi
n(1
914–
1915
,192
2–19
30),
19To
n(1
926–
1930
)S
elec
ted
orig
inal
reco
rds:
Wie
cher
t(19
30–1
968)
,G
alit
zin
(193
0–19
75),
19To
n(1
930–
1968
),M
aink
a(1
921–
1924
)
Upp
sala
,Sw
eden
Wie
cher
t(H
)D
epar
tmen
tof
Ear
thS
cien
ces,
Upp
sala
Uni
vers
ity
ww
w.g
eofy
s.uu
.se
Ver
yco
mpl
ete
Wie
cher
tcol
lect
ion
Zik
awei
,Chi
naN
ore
cord
Sha
ngha
iSei
smol
ogic
alB
urea
u
Ucc
le,B
elgi
umN
ore
cord
Obs
erva
toir
eR
oyal
deB
elgi
que
Dr.
Thi
erry
Cam
elbe
eck
T.C
amel
beec
k@om
a.be
seis
mol
ogie
.om
a.be
Fair
lyco
mpl
ete
Wie
cher
t:19
10–1
949,
Gal
itzi
n(H
):19
11–1
950
Wil
lip-
Som
vill
e(Z
):19
30–1
950
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS
374 H. Kanamori, L. Rivera and W.H.K. Lee
Tab
leC
1.(C
onti
nued
).
Sta
tion
1907
Sum
atra
reco
rdIn
stit
utio
nN
ame
–e-
mai
l–U
RL
Not
e
Bat
avia
,Ind
ones
iaN
ore
cord
Met
eoro
logi
cala
ndG
eoph
ysic
alA
genc
yw
ww
.bm
g.go
.id
Man
ila,
Phi
lipp
ines
No
reco
rdM
anil
aO
bser
vato
ryD
r.Fr
.Seg
ioS
.Su,
SJ
su@
obse
rvat
ory.
phw
ww
.obs
erva
tory
.ph
All
the
seis
mol
ogic
alre
cord
sw
ere
dest
roye
ddu
ring
Wor
ldW
arII
San
Fern
ando
,Spa
inM
ilne
publ
ishe
dco
pyR
ealI
nsti
tuto
yO
bser
vato
rio
dela
Arm
ada
ww
w.r
oa.e
s
Ebr
o,S
pain
No
reco
rdO
bser
vato
rio
del’
Ebr
ew
ww
.obs
ebre
.es
Ver
yco
mpl
ete
coll
ecti
onV
icen
tini
:190
5–19
36,G
rabl
ovit
z:19
05–1
918,
Mai
nka:
1916
–197
0
De
Bil
tt,t
heN
ethe
rlan
dsN
ore
cord
The
Roy
alN
ethe
rlan
dsM
eteo
rolo
gica
lIns
titu
teD
r.B
erna
rdD
ost
Ber
nard
.Dos
t@kn
mi.n
lw
ww
.knm
i.nl
Ver
yco
mpl
ete
coll
ecti
onof
Gal
itzi
nre
cord
ssi
nce
1912
(2ho
r.,+Z
sinc
e19
22)
Fair
lyco
mpl
ete
coll
ecti
on(1
908–
1913
and
1939
–195
8)of
Wie
cher
t(20
0kg
),an
dle
ssco
mpl
ete
(190
8–19
57)
coll
ecti
onfo
rth
eB
osch
inst
rum
ent
Chr
istc
hurc
h,N
ewZ
eala
ndN
ore
cord
Inst
itut
eof
Geo
logi
cala
ndN
ucle
arS
cien
ces
Dr.
Bri
anFe
rris
B.F
erri
s@gn
s.cr
i.nz
ww
w.g
ns.c
ri.n
z
Nea
rly
com
plet
eco
llec
tion
ofC
hris
tchu
rch
Mil
nere
cord
sfo
r19
02to
1932
,and
Gal
itzi
nZ
NE
from
1930
to19
57
Tbi
lisi
,Geo
rgia
No
reco
rdS
eism
icM
onit
orin
gC
entr
eof
Geo
rgia
seis
mo.
ilia
uni.e
du.g
e
Sam
oa,S
amoa
Har
dly
read
able
Inst
itut
fur
Geo
phys
ik,U
nive
rsit
atG
otti
ngen
ww
w.g
eo.p
hysi
k.un
i-go
etti
ngen
.de
Pre
serv
edin
Got
ting
enW
iech
ert:
16.1
2.19
02to
1.6.
1913
C© 2010 The Authors, GJI, 183, 358–374
Journal compilation C© 2010 RAS